AI mock interviewer for technical candidates that defends rubric-based reasoning scores

Who feels it

Software/technical job candidates in an active search: bootcamp grads, new-grad CS applicants, and employed engineers doing 4-10 loops over 6-12 weeks. Secondary, unverified in this block: hiring managers and recruiters who run those loops and want consistent rubric scores.

How often

Episodic, not recurring. The evidence block contains no usage-frequency data at all. Inference (stated as assumption, not evidence): heavy use for 3-8 weeks during a job search, then zero — which is exactly the churn shape that kills consumer prep subscriptions.

Why current fixes fail

The break happens at the feedback step, not the question step. A candidate can get infinite questions free (LeetCode-style banks, past interview write-ups, ChatGPT prompts), and can get a human peer to run a mock over Zoom/Meet/Teams. What neither produces is a defensible score on the reasoning path: the peer mock ends with 'that felt fine, maybe be more structured'; a generic LLM chat agrees with whatever the candidate types and never pushes on a vague answer; a static rubric PDF is not applied to what the candidate actually said. So the candidate finishes a practice session without knowing which specific assumption they failed to defend, and repeats the same failure in the real loop. The signals in this block describe exactly that gap — adaptive follow-ups that 'push on vague answers' (S-530) and scoring 'the reasoning behind each step against locked, problem-specific rubrics' (S-2041) — but they describe it as vendor launch positioning, not as candidates narrating their own week.

5 references from 2 signals · evaluation written Aug 11, 2026.